Papers by Yozen Liu
Threshold Differential Attention for Sink-Free, Ultra-Sparse, and Non-Dispersive Language Modeling (2026.acl-long)
Copied to clipboard
| Challenge: | a strict sum-to-one constraint forces attention sinks on irrelevant tokens, while probability mass disperses as sequence lengths increase. |
| Approach: | They propose a sink-free attention mechanism that achieves ultra-sparsity and improved robustness at longer sequence lengths without the computational overhead of projection methods. |
| Outcome: | The proposed mechanism produces >99 % exact zeros and eliminates attention sinks while maintaining competitive performance on standard and long-context benchmarks. |
Hierarchical Token Prepending: Enhancing Information Flow in Decoder-based LLM Embeddings (2026.acl-long)
Copied to clipboard
Xueying Ding, Xingyue Huang, Mingxuan Ju, Liam Collins, Yozen Liu, Leman Akoglu, Neil Shah, Tong Zhao
| Challenge: | Large language models produce powerful text embeddings, but their causal attention mechanism restricts the flow of information from later to earlier tokens, harming performance. |
| Approach: | They propose a method that prepending a single summary token to reduce attention-level compression by partitioning the input into blocks and prepending blocks to subsequent blocks. |
| Outcome: | The proposed method achieves consistent performance gains across 11 retrieval datasets and 30 general embedding benchmarks. |